1. Field of the Invention
The present invention relates to measuring a similarity index between captured images.
2. Description of the Related Art
With the widespread use of digital cameras and digital video cameras, captured static images or moving images are treated as electronic data. Further, cameras with pan heads, typified by WebView™ products, that can control moving images from remote locations or under automatic control are used.
Attribute information of captured images generally includes dates, times, shutter speeds, and exposures. In the case of controllable cameras, such information may include pan angles, tilt angles, angles of view (zoom angles), and even various control information or supplementary information (e.g. the temperature or weather conditions at capturing time).
Moreover, surveillance systems having the above-mentioned cameras are in use and may store captured images for later use. Since an enormous number of captured images is generally stored, especially in surveillance systems, it is necessary to reduce the enormous number of such images down to essential images. Hitherto, image retrieval using attribute information, such as the capture date and time, has been performed.
There are image retrieval techniques according to the similarity index between images, for example, disclosed in Japanese Patent Laid-Open Nos. 2000-235637 and 2000-235638 (hereinafter referred to as References 1 and 2, respectively). Using the similarity index between images allows, for example, retrieval of an image that is similar to a particular image or detecting a scene change in a TV program to find a part where images vary in moving images.
However, such retrieval has a problem in that a retrieved image from an enormous number of images is not always the exact image that meets the needs of searchers. In particular, measurement of the similarity index depending on captured images themselves is insufficient for accurate measurement since all capturing conditions are not considered.
Capturing with a movable camera or the like may produce images having different angles of view according to control, so that the captured images may include images that happen to show similarity, such as images whose captured objects are themselves different but have similar color or shape. Therefore, retrieval depending solely on the similarity index between images has problems.
Further, capturing with a movable camera or the like may include a case where a captured object of interest, which is specified as a rectangular region in a selected image, does not exist in other images due to the different angles of view or a case where the captured object exists at different locations in the images.
In view of these circumstances, it is an object of the present invention to provide a method for processing images that reduces an enormous number of images down to images suitable for measuring the similarity index and that measures the similarity index between the reduced number of images more effectively.